Process
How we work
MagicPlug builds software in short Agile iterations with AI coding agents. You approve each spec, and an engineer approves every merge.
We build software in short Agile iterations, with AI coding agents in each one. Our engineers write the specs and decide what merges.
The delivery loop
Steps 2 to 5 repeat in each iteration. At the end of each iteration, you see working software and we plan the next one with you.
- Discover. We learn your goals and users, and AI summarises your documents. You approve the scope, and we set up CI and rules files.
- Specify. We write a short spec with acceptance criteria for each feature, plus a design or prototype when it needs one. AI checks the spec for gaps, and you approve it before any code starts.
- Build. Coding agents draft the code and its tests in a sandbox, from the spec and the rules files. An engineer steers the agents and rejects work that misses the spec.
- Review. CI runs the tests, plus the eval set for AI features, and an AI reviewer flags bugs and security issues. Then an engineer reads the diff, runs the change and approves it or sends it back.
- Release. You try each change on a preview link or a test build. We release the work that you accept.
- Support. We monitor the live system and fix issues. AI helps us read logs and alerts, and an engineer reviews each fix.
What you own
You own all code, specs, docs, rules files, tests and eval sets. They are in your repositories from the first commit. The code runs without a MagicPlug account or any AI coding tool.
The rules files and docs tell any engineer or agent how your codebase works. We can also move your team to this workflow with AI-Native Development.
How we control AI work
- Branch protection lets only an engineer approve a merge, so no agent can merge its own work.
- Agents work in a sandbox with test data and never get production secrets.
- Agents use only the data, tools and MCP servers that you approve.
- Our AI coding tools run on business plans whose terms keep your code out of model training. We can also run models in your own cloud account.
- Each AI feature that we build gets an eval set, so you see its test scores before release.
How we start
We start with one small milestone, such as one feature in one repository. You see the full loop on real work before you agree to more. Tell us about your project.
Start a project
Tell us about your project.
Send us a short note about what you want to build. We reply within one working day.
Contact usEmail
hello@magicplug.techPhone and WhatsApp
+91-7837770661Message on WhatsApp